PostProcessingUtils: Add method for normal reconstruction.

[Mugen87]

Related issue: -

Description

This PR introduces PostProcessingUtils.getNormalFromDepth() which can be used to compute normals from depth. This can be useful if MRT is too costly for some reason (e.g. bandwidth restrictions).

I've updated the AO modules to use this new code path if no normal render target is available.

I want to make clear that reconstructed normals are slightly different compared to our current normal render target since it represents the actual flat surface of the geometry. This technique is inadequate for stuff like SSR since you end up with a flat-shaded, facetted look. However, it is actually the better pick for AO which is explained in https://wickedengine.net/2019/09/improved-normal-reconstruction-from-depth/comment-page-1/.

@sunag The quality difference is barely visible in webgpu_postprocessing_ao but we should keep in mind this issue when using transformedNormalView for normal texture input.

[github-actions]

📦 Bundle size

Full ESM build, minified and gzipped.

	Before	After	Diff
WebGL	692.71 171.44	692.71 171.44	+0 B +0 B
WebGPU	817.37 220.29	818.16 220.56	+791 B +262 B
WebGPU Nodes	816.88 220.17	817.67 220.43	+791 B +262 B

🌳 Bundle size after tree-shaking

Minimal build including a renderer, camera, empty scene, and dependencies.

	Before	After	Diff
WebGL	464.93 112.2	464.93 112.2	+0 B +0 B
WebGPU	539.13 145.56	539.13 145.56	+0 B +0 B
WebGPU Nodes	495.16 135.43	495.16 135.43	+0 B +0 B

[Mugen87]

@sunag It seems there is an issue when using depth textures with textureLoad() and the WebGL backend. There are no issues in WebGPU. The following shader errors occur:

ERROR: 0:149: '=' : dimension mismatch
ERROR: 0:149: 'assign' : cannot convert from 'highp 4-component vector of float' to 'highp float'

Live example: https://jsfiddle.net/1tkpy5ja/

I think this is related to the fact that texelFetch always returns a four-component vector. textureLoad is more flexibel and can also return f32 values. When I read the code correctly, this difference between GLSL and WGSL isn't honored yet.